The Stem Quality Prediction Dataset comprises 3,014 samples of roundwood, categorized into 1,450 class B, 1,450 class C, and 114 class D samples. For each sample there are two images: a top view image of the stem and one of the cross-section. It consists of 2,411 images (80%) for training and 603 images (20%) for testing, with an equal distribution of classes in both splits. The labels only reflect the trunk features — excluding cross-sectional features. Each sample is saved along with its corresponding cross-sectional image and a log file. The log file follows this structure: log; protocol number; stem length [cm]; diameter D1 base [mm]; diameter D2 middle [mm]; diameter D3 top [mm]; taper [mm/m]; species; quality.